search

breagen / MachSuite

Benchmarks for Accelerator Design and Customized Architectures
http://www.eecs.harvard.edu/~reagen/papers/machsuite.pdf
Other
114 stars 47 forks source link

Backprop generator is missing data files #6

Open rdadolf opened 8 years ago

rdadolf commented 8 years ago

Our old data generator used rather lengthy headers for specifying training data and labels. We removed them, but the generator code still references their variables.

The right thing to do is create a synthetic training set by adding noise to a computable function. This would fix the undefined variable problems, and it would also enable backprop/backprop to work properly when the input_dimension and possible_outputs parameters are changed.

johnnysDeu commented 7 years ago

hello

is there any chance that you update a working version of backprop? i am a student and i need it for a project thanks a lot

rdadolf commented 7 years ago

Hi @johnnysDeu. Just to be clear about this, backprop itself does work, it's just the generator which is affected by this. Moreover, it's not too terribly tough to modify backprop to use a different neural net topology, the bug just means that MachSuite won't auto-generate a set of verification outputs for you to check against.

At the moment, this isn't on the top of our priority list, but if it's becoming a roadblock for more users, then we can revisit it.

hkultala commented 5 years ago

I am also affected by this. I need to do some porting because my platform does not support file IO, so I cannot use the ready data files, I need to generate the data into headers and link it inside the binary.